[Triton] Make the Triton device guards accelerator-agnostic #8868
VenusTZZ
marked this pull request as draft 7 days ago
VenusTZZ
force pushed
from
77098299
to
5cc7afe9
7 days ago
VenusTZZ
marked this pull request as ready for review 7 days ago
Make the Triton device guards accelerator-agnostic
9488bd68
VenusTZZ
force pushed
from
5cc7afe9
to
9488bd68
7 days ago
Check NPU device availability before opening the Triton gate
37011b0d
VenusTZZ
marked this pull request as draft 6 days ago
Merge branch 'deepspeedai:master' into npu-triton-support
509605a5
Gate every Triton kernel call site behind a runtime probe
8985183b
Harden the Triton kernel probes
d167b4b4
Add unit tests for the per-kernel Triton probe gate
acfbf62d
Check the accelerator device generator in the probe RNG test
1371124b
Draw probe inputs deterministically and harden the probe gate
f803305f
Answer the Triton gather gate from is_triton_supported
a2a9888a
VenusTZZ
marked this pull request as ready for review 5 days ago
Dispatch triton_ops kernels through per-hardware subpackages
36e1b910
VenusTZZ
marked this pull request as draft 5 days ago
Dispatch every training-path Triton kernel entry through an annotation
542efb94
Add FLA-style Ascend variants for the Triton kernels FLA implements
eb817bca
VenusTZZ
changed the title Make the Triton device guards accelerator-agnostic [Triton] Make the Triton device guards accelerator-agnostic 2 days ago
VenusTZZ
marked this pull request as ready for review 1 day ago
Assignees
No one assigned
Login to write a write a comment.
Login via GitHub